Hi there,
I think this may be inappropriate for this list as I
searched the archive and all the messages of this type were pointed to e-mail
to '
I’m trying to fit a Hierarchical Model that performs a
meta-analysis across study type but accounting for study type. There are
seven studies, 2 that are randomized and 5 that are not. The data in each
of the studies is the observed deaths (rmortRT) out of the total number
patients (nmortRT). The study design information is held in the variable
DesignRT. DesignRT=1 is a randomized design and DesignRT=2 is a
non-randomized design.
Each study design type is expected to have a mean death rate
after logit transformation, muRT[DesignRT[i]]. The overall population
mean death rate (combining over study type) after logit transformation is
munotRT.
When I compile the model and the data, I get a multiple
definitions of node muRT[2] message. Why isn’t this working? This
project is due next week so any help that you could give me in the next day or
so would be greatly appreciated. I’m sort of at a standstill.
Take care,
Teresa
model{ for (i in
1:Nstudy){
rmortRT[i] ~ dbin(pRT[i], nmortRT[i])
logit(pRT[i]) <- RT[i]
RT[i] ~ dnorm(muRT[DesignRT[i]], tauRT[DesignRT[i]])
useless[i] <- StudyRT[i] + StentNumRT[i]+ gpiiaiiibRT[i]
muRT[DesignRT[i]] ~ dnorm(munotRT, taunotRT)
}
tauRT[1] <- pow(sigmaRT[1], -2)
tauRT[2] <- pow(sigmaRT[2], -2)
sigmaRT[1] ~ dunif(0,100)
sigmaRT[2] ~ dunif(0, 100)
ppostRT[1] <-
exp(muRT[DesignRT[1]])/(1 + exp(muRT[DesignRT[1]]))
ppostRT[2] <-
exp(muRT[DesignRT[2]])/(1 + exp(muRT[DesignRT[2]]))
munotRT ~ dnorm(0, 0001)
taunotRT <-pow(sigmanotRT, -2)
sigmanotRT ~ dunif(0,100)
ppostRTnot <- exp(munotRT)/(1 +
exp(munotRT))
}
#DATA AJ STUDIES
list(Nstudy=7)
StudyRT[]
DesignRT[]
StentNumRT[]
gpiiaiiibRT[]
nmortRT[]
rmortRT[]
1
1
1
1
50 0
2
2
0.913
0.87 46
2
3
2
NA
NA
80 3
4
2
0.39
0
31 0
5
2
0.67
0.262
70 5
6
1
0.93
0.95
240 11
7
2
1
0.88
52 1
END
Teresa Nelson, M.S.
Principal Statistician
Princeton Reimbursement Group
Phone: (952) 345-6413
Fax: (715) 755-2767
e-mail: [log in to unmask]